falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
65064 Commits
32 Branches
824MB
Tree: f3158c3f29
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f3158c3f29'
${ noResults }
FFmpeg/libswscale/swscale.c

1089 lines
42KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2011 Michael Niedermayer <michaelni@gmx.at>

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * FFmpeg is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with FFmpeg; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. #include <inttypes.h>

  22. #include <math.h>

  23. #include <stdio.h>

  24. #include <string.h>

  25. 

  26. #include "libavutil/avassert.h"

  27. #include "libavutil/avutil.h"

  28. #include "libavutil/bswap.h"

  29. #include "libavutil/cpu.h"

  30. #include "libavutil/intreadwrite.h"

  31. #include "libavutil/mathematics.h"

  32. #include "libavutil/pixdesc.h"

  33. #include "config.h"

  34. #include "rgb2rgb.h"

  35. #include "swscale_internal.h"

  36. #include "swscale.h"

  37. 

  38. DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = {

  39.     {  36, 68,  60, 92,  34, 66,  58, 90, },

  40.     { 100,  4, 124, 28,  98,  2, 122, 26, },

  41.     {  52, 84,  44, 76,  50, 82,  42, 74, },

  42.     { 116, 20, 108, 12, 114, 18, 106, 10, },

  43.     {  32, 64,  56, 88,  38, 70,  62, 94, },

  44.     {  96,  0, 120, 24, 102,  6, 126, 30, },

  45.     {  48, 80,  40, 72,  54, 86,  46, 78, },

  46.     { 112, 16, 104,  8, 118, 22, 110, 14, },

  47.     {  36, 68,  60, 92,  34, 66,  58, 90, },

  48. };

  49. 

  50. DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {

  51.     64, 64, 64, 64, 64, 64, 64, 64

  52. };

  53. 

  54. static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,

  55.                                        int height, int y, uint8_t val)

  56. {

  57.     int i;

  58.     uint8_t *ptr = plane + stride * y;

  59.     for (i = 0; i < height; i++) {

  60.         memset(ptr, val, width);

  61.         ptr += stride;

  62.     }

  63. }

  64. 

  65. static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,

  66.                            const uint8_t *_src, const int16_t *filter,

  67.                            const int32_t *filterPos, int filterSize)

  68. {

  69.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);

  70.     int i;

  71.     int32_t *dst        = (int32_t *) _dst;

  72.     const uint16_t *src = (const uint16_t *) _src;

  73.     int bits            = desc->comp[0].depth_minus1;

  74.     int sh              = bits - 4;

  75. 

  76.     if((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth_minus1<15)

  77.         sh= 9;

  78. 

  79.     for (i = 0; i < dstW; i++) {

  80.         int j;

  81.         int srcPos = filterPos[i];

  82.         int val    = 0;

  83. 

  84.         for (j = 0; j < filterSize; j++) {

  85.             val += src[srcPos + j] * filter[filterSize * i + j];

  86.         }

  87.         // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit

  88.         dst[i] = FFMIN(val >> sh, (1 << 19) - 1);

  89.     }

  90. }

  91. 

  92. static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,

  93.                            const uint8_t *_src, const int16_t *filter,

  94.                            const int32_t *filterPos, int filterSize)

  95. {

  96.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);

  97.     int i;

  98.     const uint16_t *src = (const uint16_t *) _src;

  99.     int sh              = desc->comp[0].depth_minus1;

  100. 

  101.     if(sh<15)

  102.         sh= isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : desc->comp[0].depth_minus1;

  103. 

  104.     for (i = 0; i < dstW; i++) {

  105.         int j;

  106.         int srcPos = filterPos[i];

  107.         int val    = 0;

  108. 

  109.         for (j = 0; j < filterSize; j++) {

  110.             val += src[srcPos + j] * filter[filterSize * i + j];

  111.         }

  112.         // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit

  113.         dst[i] = FFMIN(val >> sh, (1 << 15) - 1);

  114.     }

  115. }

  116. 

  117. // bilinear / bicubic scaling

  118. static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,

  119.                           const uint8_t *src, const int16_t *filter,

  120.                           const int32_t *filterPos, int filterSize)

  121. {

  122.     int i;

  123.     for (i = 0; i < dstW; i++) {

  124.         int j;

  125.         int srcPos = filterPos[i];

  126.         int val    = 0;

  127.         for (j = 0; j < filterSize; j++) {

  128.             val += ((int)src[srcPos + j]) * filter[filterSize * i + j];

  129.         }

  130.         dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...

  131.     }

  132. }

  133. 

  134. static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,

  135.                           const uint8_t *src, const int16_t *filter,

  136.                           const int32_t *filterPos, int filterSize)

  137. {

  138.     int i;

  139.     int32_t *dst = (int32_t *) _dst;

  140.     for (i = 0; i < dstW; i++) {

  141.         int j;

  142.         int srcPos = filterPos[i];

  143.         int val    = 0;

  144.         for (j = 0; j < filterSize; j++) {

  145.             val += ((int)src[srcPos + j]) * filter[filterSize * i + j];

  146.         }

  147.         dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...

  148.     }

  149. }

  150. 

  151. // FIXME all pal and rgb srcFormats could do this conversion as well

  152. // FIXME all scalers more complex than bilinear could do half of this transform

  153. static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)

  154. {

  155.     int i;

  156.     for (i = 0; i < width; i++) {

  157.         dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264

  158.         dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264

  159.     }

  160. }

  161. 

  162. static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)

  163. {

  164.     int i;

  165.     for (i = 0; i < width; i++) {

  166.         dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469

  167.         dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469

  168.     }

  169. }

  170. 

  171. static void lumRangeToJpeg_c(int16_t *dst, int width)

  172. {

  173.     int i;

  174.     for (i = 0; i < width; i++)

  175.         dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;

  176. }

  177. 

  178. static void lumRangeFromJpeg_c(int16_t *dst, int width)

  179. {

  180.     int i;

  181.     for (i = 0; i < width; i++)

  182.         dst[i] = (dst[i] * 14071 + 33561947) >> 14;

  183. }

  184. 

  185. static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)

  186. {

  187.     int i;

  188.     int32_t *dstU = (int32_t *) _dstU;

  189.     int32_t *dstV = (int32_t *) _dstV;

  190.     for (i = 0; i < width; i++) {

  191.         dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264

  192.         dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264

  193.     }

  194. }

  195. 

  196. static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)

  197. {

  198.     int i;

  199.     int32_t *dstU = (int32_t *) _dstU;

  200.     int32_t *dstV = (int32_t *) _dstV;

  201.     for (i = 0; i < width; i++) {

  202.         dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469

  203.         dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469

  204.     }

  205. }

  206. 

  207. static void lumRangeToJpeg16_c(int16_t *_dst, int width)

  208. {

  209.     int i;

  210.     int32_t *dst = (int32_t *) _dst;

  211.     for (i = 0; i < width; i++) {

  212.         dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12;

  213.     }

  214. }

  215. 

  216. static void lumRangeFromJpeg16_c(int16_t *_dst, int width)

  217. {

  218.     int i;

  219.     int32_t *dst = (int32_t *) _dst;

  220.     for (i = 0; i < width; i++)

  221.         dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12;

  222. }

  223. 

  224. // *** horizontal scale Y line to temp buffer

  225. static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,

  226.                                      const uint8_t *src_in[4],

  227.                                      int srcW, int xInc,

  228.                                      const int16_t *hLumFilter,

  229.                                      const int32_t *hLumFilterPos,

  230.                                      int hLumFilterSize,

  231.                                      uint8_t *formatConvBuffer,

  232.                                      uint32_t *pal, int isAlpha)

  233. {

  234.     void (*toYV12)(uint8_t *, const uint8_t *, const uint8_t *, const uint8_t *, int, uint32_t *) =

  235.         isAlpha ? c->alpToYV12 : c->lumToYV12;

  236.     void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;

  237.     const uint8_t *src = src_in[isAlpha ? 3 : 0];

  238. 

  239.     if (toYV12) {

  240.         toYV12(formatConvBuffer, src, src_in[1], src_in[2], srcW, pal);

  241.         src = formatConvBuffer;

  242.     } else if (c->readLumPlanar && !isAlpha) {

  243.         c->readLumPlanar(formatConvBuffer, src_in, srcW, c->input_rgb2yuv_table);

  244.         src = formatConvBuffer;

  245.     } else if (c->readAlpPlanar && isAlpha) {

  246.         c->readAlpPlanar(formatConvBuffer, src_in, srcW, NULL);

  247.         src = formatConvBuffer;

  248.     }

  249. 

  250.     if (!c->hyscale_fast) {

  251.         c->hyScale(c, dst, dstWidth, src, hLumFilter,

  252.                    hLumFilterPos, hLumFilterSize);

  253.     } else { // fast bilinear upscale / crap downscale

  254.         c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);

  255.     }

  256. 

  257.     if (convertRange)

  258.         convertRange(dst, dstWidth);

  259. }

  260. 

  261. static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,

  262.                                      int16_t *dst2, int dstWidth,

  263.                                      const uint8_t *src_in[4],

  264.                                      int srcW, int xInc,

  265.                                      const int16_t *hChrFilter,

  266.                                      const int32_t *hChrFilterPos,

  267.                                      int hChrFilterSize,

  268.                                      uint8_t *formatConvBuffer, uint32_t *pal)

  269. {

  270.     const uint8_t *src1 = src_in[1], *src2 = src_in[2];

  271.     if (c->chrToYV12) {

  272.         uint8_t *buf2 = formatConvBuffer +

  273.                         FFALIGN(srcW*2+78, 16);

  274.         c->chrToYV12(formatConvBuffer, buf2, src_in[0], src1, src2, srcW, pal);

  275.         src1= formatConvBuffer;

  276.         src2= buf2;

  277.     } else if (c->readChrPlanar) {

  278.         uint8_t *buf2 = formatConvBuffer +

  279.                         FFALIGN(srcW*2+78, 16);

  280.         c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW, c->input_rgb2yuv_table);

  281.         src1 = formatConvBuffer;

  282.         src2 = buf2;

  283.     }

  284. 

  285.     if (!c->hcscale_fast) {

  286.         c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);

  287.         c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);

  288.     } else { // fast bilinear upscale / crap downscale

  289.         c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);

  290.     }

  291. 

  292.     if (c->chrConvertRange)

  293.         c->chrConvertRange(dst1, dst2, dstWidth);

  294. }

  295. 

  296. #define DEBUG_SWSCALE_BUFFERS 0

  297. #define DEBUG_BUFFERS(...)                      \

  298.     if (DEBUG_SWSCALE_BUFFERS)                  \

  299.         av_log(c, AV_LOG_DEBUG, __VA_ARGS__)

  300. 

  301. static int swscale(SwsContext *c, const uint8_t *src[],

  302.                    int srcStride[], int srcSliceY,

  303.                    int srcSliceH, uint8_t *dst[], int dstStride[])

  304. {

  305.     /* load a few things into local vars to make the code more readable?

  306.      * and faster */

  307.     const int srcW                   = c->srcW;

  308.     const int dstW                   = c->dstW;

  309.     const int dstH                   = c->dstH;

  310.     const int chrDstW                = c->chrDstW;

  311.     const int chrSrcW                = c->chrSrcW;

  312.     const int lumXInc                = c->lumXInc;

  313.     const int chrXInc                = c->chrXInc;

  314.     const enum AVPixelFormat dstFormat = c->dstFormat;

  315.     const int flags                  = c->flags;

  316.     int32_t *vLumFilterPos           = c->vLumFilterPos;

  317.     int32_t *vChrFilterPos           = c->vChrFilterPos;

  318.     int32_t *hLumFilterPos           = c->hLumFilterPos;

  319.     int32_t *hChrFilterPos           = c->hChrFilterPos;

  320.     int16_t *hLumFilter              = c->hLumFilter;

  321.     int16_t *hChrFilter              = c->hChrFilter;

  322.     int32_t *lumMmxFilter            = c->lumMmxFilter;

  323.     int32_t *chrMmxFilter            = c->chrMmxFilter;

  324.     const int vLumFilterSize         = c->vLumFilterSize;

  325.     const int vChrFilterSize         = c->vChrFilterSize;

  326.     const int hLumFilterSize         = c->hLumFilterSize;

  327.     const int hChrFilterSize         = c->hChrFilterSize;

  328.     int16_t **lumPixBuf              = c->lumPixBuf;

  329.     int16_t **chrUPixBuf             = c->chrUPixBuf;

  330.     int16_t **chrVPixBuf             = c->chrVPixBuf;

  331.     int16_t **alpPixBuf              = c->alpPixBuf;

  332.     const int vLumBufSize            = c->vLumBufSize;

  333.     const int vChrBufSize            = c->vChrBufSize;

  334.     uint8_t *formatConvBuffer        = c->formatConvBuffer;

  335.     uint32_t *pal                    = c->pal_yuv;

  336.     yuv2planar1_fn yuv2plane1        = c->yuv2plane1;

  337.     yuv2planarX_fn yuv2planeX        = c->yuv2planeX;

  338.     yuv2interleavedX_fn yuv2nv12cX   = c->yuv2nv12cX;

  339.     yuv2packed1_fn yuv2packed1       = c->yuv2packed1;

  340.     yuv2packed2_fn yuv2packed2       = c->yuv2packed2;

  341.     yuv2packedX_fn yuv2packedX       = c->yuv2packedX;

  342.     yuv2anyX_fn yuv2anyX             = c->yuv2anyX;

  343.     const int chrSrcSliceY           =                srcSliceY >> c->chrSrcVSubSample;

  344.     const int chrSrcSliceH           = FF_CEIL_RSHIFT(srcSliceH,   c->chrSrcVSubSample);

  345.     int should_dither                = is9_OR_10BPS(c->srcFormat) ||

  346.                                        is16BPS(c->srcFormat);

  347.     int lastDstY;

  348. 

  349.     /* vars which will change and which we need to store back in the context */

  350.     int dstY         = c->dstY;

  351.     int lumBufIndex  = c->lumBufIndex;

  352.     int chrBufIndex  = c->chrBufIndex;

  353.     int lastInLumBuf = c->lastInLumBuf;

  354.     int lastInChrBuf = c->lastInChrBuf;

  355. 

  356.     if (!usePal(c->srcFormat)) {

  357.         pal = c->input_rgb2yuv_table;

  358.     }

  359. 

  360.     if (isPacked(c->srcFormat)) {

  361.         src[0] =

  362.         src[1] =

  363.         src[2] =

  364.         src[3] = src[0];

  365.         srcStride[0] =

  366.         srcStride[1] =

  367.         srcStride[2] =

  368.         srcStride[3] = srcStride[0];

  369.     }

  370.     srcStride[1] <<= c->vChrDrop;

  371.     srcStride[2] <<= c->vChrDrop;

  372. 

  373.     DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",

  374.                   src[0], srcStride[0], src[1], srcStride[1],

  375.                   src[2], srcStride[2], src[3], srcStride[3],

  376.                   dst[0], dstStride[0], dst[1], dstStride[1],

  377.                   dst[2], dstStride[2], dst[3], dstStride[3]);

  378.     DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",

  379.                   srcSliceY, srcSliceH, dstY, dstH);

  380.     DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",

  381.                   vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);

  382. 

  383.     if (dstStride[0]&15 || dstStride[1]&15 ||

  384.         dstStride[2]&15 || dstStride[3]&15) {

  385.         static int warnedAlready = 0; // FIXME maybe move this into the context

  386.         if (flags & SWS_PRINT_INFO && !warnedAlready) {

  387.             av_log(c, AV_LOG_WARNING,

  388.                    "Warning: dstStride is not aligned!\n"

  389.                    "         ->cannot do aligned memory accesses anymore\n");

  390.             warnedAlready = 1;

  391.         }

  392.     }

  393. 

  394.     if (   (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15

  395.         || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15

  396.         || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15

  397.         || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15

  398.     ) {

  399.         static int warnedAlready=0;

  400.         int cpu_flags = av_get_cpu_flags();

  401.         if (HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && !warnedAlready){

  402.             av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speedloss\n");

  403.             warnedAlready=1;

  404.         }

  405.     }

  406. 

  407.     /* Note the user might start scaling the picture in the middle so this

  408.      * will not get executed. This is not really intended but works

  409.      * currently, so people might do it. */

  410.     if (srcSliceY == 0) {

  411.         lumBufIndex  = -1;

  412.         chrBufIndex  = -1;

  413.         dstY         = 0;

  414.         lastInLumBuf = -1;

  415.         lastInChrBuf = -1;

  416.     }

  417. 

  418.     if (!should_dither) {

  419.         c->chrDither8 = c->lumDither8 = sws_pb_64;

  420.     }

  421.     lastDstY = dstY;

  422. 

  423.     for (; dstY < dstH; dstY++) {

  424.         const int chrDstY = dstY >> c->chrDstVSubSample;

  425.         uint8_t *dest[4]  = {

  426.             dst[0] + dstStride[0] * dstY,

  427.             dst[1] + dstStride[1] * chrDstY,

  428.             dst[2] + dstStride[2] * chrDstY,

  429.             (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,

  430.         };

  431.         int use_mmx_vfilter= c->use_mmx_vfilter;

  432. 

  433.         // First line needed as input

  434.         const int firstLumSrcY  = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);

  435.         const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);

  436.         // First line needed as input

  437.         const int firstChrSrcY  = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);

  438. 

  439.         // Last line needed as input

  440.         int lastLumSrcY  = FFMIN(c->srcH,    firstLumSrcY  + vLumFilterSize) - 1;

  441.         int lastLumSrcY2 = FFMIN(c->srcH,    firstLumSrcY2 + vLumFilterSize) - 1;

  442.         int lastChrSrcY  = FFMIN(c->chrSrcH, firstChrSrcY  + vChrFilterSize) - 1;

  443.         int enough_lines;

  444. 

  445.         // handle holes (FAST_BILINEAR & weird filters)

  446.         if (firstLumSrcY > lastInLumBuf)

  447.             lastInLumBuf = firstLumSrcY - 1;

  448.         if (firstChrSrcY > lastInChrBuf)

  449.             lastInChrBuf = firstChrSrcY - 1;

  450.         av_assert0(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);

  451.         av_assert0(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);

  452. 

  453.         DEBUG_BUFFERS("dstY: %d\n", dstY);

  454.         DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",

  455.                       firstLumSrcY, lastLumSrcY, lastInLumBuf);

  456.         DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",

  457.                       firstChrSrcY, lastChrSrcY, lastInChrBuf);

  458. 

  459.         // Do we have enough lines in this slice to output the dstY line

  460.         enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&

  461.                        lastChrSrcY < FF_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);

  462. 

  463.         if (!enough_lines) {

  464.             lastLumSrcY = srcSliceY + srcSliceH - 1;

  465.             lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;

  466.             DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",

  467.                           lastLumSrcY, lastChrSrcY);

  468.         }

  469. 

  470.         // Do horizontal scaling

  471.         while (lastInLumBuf < lastLumSrcY) {

  472.             const uint8_t *src1[4] = {

  473.                 src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],

  474.                 src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],

  475.                 src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],

  476.                 src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],

  477.             };

  478.             lumBufIndex++;

  479.             av_assert0(lumBufIndex < 2 * vLumBufSize);

  480.             av_assert0(lastInLumBuf + 1 - srcSliceY < srcSliceH);

  481.             av_assert0(lastInLumBuf + 1 - srcSliceY >= 0);

  482.             hyscale(c, lumPixBuf[lumBufIndex], dstW, src1, srcW, lumXInc,

  483.                     hLumFilter, hLumFilterPos, hLumFilterSize,

  484.                     formatConvBuffer, pal, 0);

  485.             if (CONFIG_SWSCALE_ALPHA && alpPixBuf)

  486.                 hyscale(c, alpPixBuf[lumBufIndex], dstW, src1, srcW,

  487.                         lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,

  488.                         formatConvBuffer, pal, 1);

  489.             lastInLumBuf++;

  490.             DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",

  491.                           lumBufIndex, lastInLumBuf);

  492.         }

  493.         while (lastInChrBuf < lastChrSrcY) {

  494.             const uint8_t *src1[4] = {

  495.                 src[0] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[0],

  496.                 src[1] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[1],

  497.                 src[2] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[2],

  498.                 src[3] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[3],

  499.             };

  500.             chrBufIndex++;

  501.             av_assert0(chrBufIndex < 2 * vChrBufSize);

  502.             av_assert0(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));

  503.             av_assert0(lastInChrBuf + 1 - chrSrcSliceY >= 0);

  504.             // FIXME replace parameters through context struct (some at least)

  505. 

  506.             if (c->needs_hcscale)

  507.                 hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],

  508.                         chrDstW, src1, chrSrcW, chrXInc,

  509.                         hChrFilter, hChrFilterPos, hChrFilterSize,

  510.                         formatConvBuffer, pal);

  511.             lastInChrBuf++;

  512.             DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",

  513.                           chrBufIndex, lastInChrBuf);

  514.         }

  515.         // wrap buf index around to stay inside the ring buffer

  516.         if (lumBufIndex >= vLumBufSize)

  517.             lumBufIndex -= vLumBufSize;

  518.         if (chrBufIndex >= vChrBufSize)

  519.             chrBufIndex -= vChrBufSize;

  520.         if (!enough_lines)

  521.             break;  // we can't output a dstY line so let's try with the next slice

  522. 

  523. #if HAVE_MMX_INLINE

  524.         updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,

  525.                               lastInLumBuf, lastInChrBuf);

  526. #endif

  527.         if (should_dither) {

  528.             c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];

  529.             c->lumDither8 = ff_dither_8x8_128[dstY    & 7];

  530.         }

  531.         if (dstY >= dstH - 2) {

  532.             /* hmm looks like we can't use MMX here without overwriting

  533.              * this array's tail */

  534.             ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,

  535.                                      &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);

  536.             use_mmx_vfilter= 0;

  537.         }

  538. 

  539.         {

  540.             const int16_t **lumSrcPtr  = (const int16_t **)(void*) lumPixBuf  + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;

  541.             const int16_t **chrUSrcPtr = (const int16_t **)(void*) chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;

  542.             const int16_t **chrVSrcPtr = (const int16_t **)(void*) chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;

  543.             const int16_t **alpSrcPtr  = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?

  544.                                          (const int16_t **)(void*) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;

  545.             int16_t *vLumFilter = c->vLumFilter;

  546.             int16_t *vChrFilter = c->vChrFilter;

  547. 

  548.             if (isPlanarYUV(dstFormat) ||

  549.                 (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like

  550.                 const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;

  551. 

  552.                 vLumFilter +=    dstY * vLumFilterSize;

  553.                 vChrFilter += chrDstY * vChrFilterSize;

  554. 

  555. //                 av_assert0(use_mmx_vfilter != (

  556. //                                yuv2planeX == yuv2planeX_10BE_c

  557. //                             || yuv2planeX == yuv2planeX_10LE_c

  558. //                             || yuv2planeX == yuv2planeX_9BE_c

  559. //                             || yuv2planeX == yuv2planeX_9LE_c

  560. //                             || yuv2planeX == yuv2planeX_16BE_c

  561. //                             || yuv2planeX == yuv2planeX_16LE_c

  562. //                             || yuv2planeX == yuv2planeX_8_c) || !ARCH_X86);

  563. 

  564.                 if(use_mmx_vfilter){

  565.                     vLumFilter= (int16_t *)c->lumMmxFilter;

  566.                     vChrFilter= (int16_t *)c->chrMmxFilter;

  567.                 }

  568. 

  569.                 if (vLumFilterSize == 1) {

  570.                     yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);

  571.                 } else {

  572.                     yuv2planeX(vLumFilter, vLumFilterSize,

  573.                                lumSrcPtr, dest[0],

  574.                                dstW, c->lumDither8, 0);

  575.                 }

  576. 

  577.                 if (!((dstY & chrSkipMask) || isGray(dstFormat))) {

  578.                     if (yuv2nv12cX) {

  579.                         yuv2nv12cX(c, vChrFilter,

  580.                                    vChrFilterSize, chrUSrcPtr, chrVSrcPtr,

  581.                                    dest[1], chrDstW);

  582.                     } else if (vChrFilterSize == 1) {

  583.                         yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);

  584.                         yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);

  585.                     } else {

  586.                         yuv2planeX(vChrFilter,

  587.                                    vChrFilterSize, chrUSrcPtr, dest[1],

  588.                                    chrDstW, c->chrDither8, 0);

  589.                         yuv2planeX(vChrFilter,

  590.                                    vChrFilterSize, chrVSrcPtr, dest[2],

  591.                                    chrDstW, c->chrDither8, use_mmx_vfilter ? (c->uv_offx2 >> 1) : 3);

  592.                     }

  593.                 }

  594. 

  595.                 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {

  596.                     if(use_mmx_vfilter){

  597.                         vLumFilter= (int16_t *)c->alpMmxFilter;

  598.                     }

  599.                     if (vLumFilterSize == 1) {

  600.                         yuv2plane1(alpSrcPtr[0], dest[3], dstW,

  601.                                    c->lumDither8, 0);

  602.                     } else {

  603.                         yuv2planeX(vLumFilter,

  604.                                    vLumFilterSize, alpSrcPtr, dest[3],

  605.                                    dstW, c->lumDither8, 0);

  606.                     }

  607.                 }

  608.             } else if (yuv2packedX) {

  609.                 av_assert1(lumSrcPtr  + vLumFilterSize - 1 < (const int16_t **)lumPixBuf  + vLumBufSize * 2);

  610.                 av_assert1(chrUSrcPtr + vChrFilterSize - 1 < (const int16_t **)chrUPixBuf + vChrBufSize * 2);

  611.                 if (c->yuv2packed1 && vLumFilterSize == 1 &&

  612.                     vChrFilterSize <= 2) { // unscaled RGB

  613.                     int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];

  614.                     yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,

  615.                                 alpPixBuf ? *alpSrcPtr : NULL,

  616.                                 dest[0], dstW, chrAlpha, dstY);

  617.                 } else if (c->yuv2packed2 && vLumFilterSize == 2 &&

  618.                            vChrFilterSize == 2) { // bilinear upscale RGB

  619.                     int lumAlpha = vLumFilter[2 * dstY + 1];

  620.                     int chrAlpha = vChrFilter[2 * dstY + 1];

  621.                     lumMmxFilter[2] =

  622.                     lumMmxFilter[3] = vLumFilter[2 * dstY]    * 0x10001;

  623.                     chrMmxFilter[2] =

  624.                     chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;

  625.                     yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,

  626.                                 alpPixBuf ? alpSrcPtr : NULL,

  627.                                 dest[0], dstW, lumAlpha, chrAlpha, dstY);

  628.                 } else { // general RGB

  629.                     yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,

  630.                                 lumSrcPtr, vLumFilterSize,

  631.                                 vChrFilter + dstY * vChrFilterSize,

  632.                                 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,

  633.                                 alpSrcPtr, dest[0], dstW, dstY);

  634.                 }

  635.             } else {

  636.                 av_assert1(!yuv2packed1 && !yuv2packed2);

  637.                 yuv2anyX(c, vLumFilter + dstY * vLumFilterSize,

  638.                          lumSrcPtr, vLumFilterSize,

  639.                          vChrFilter + dstY * vChrFilterSize,

  640.                          chrUSrcPtr, chrVSrcPtr, vChrFilterSize,

  641.                          alpSrcPtr, dest, dstW, dstY);

  642.             }

  643.         }

  644.     }

  645.     if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {

  646.         int length = dstW;

  647.         int height = dstY - lastDstY;

  648. 

  649.         if (is16BPS(dstFormat) || isNBPS(dstFormat)) {

  650.             const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);

  651.             fillPlane16(dst[3], dstStride[3], length, height, lastDstY,

  652.                     1, desc->comp[3].depth_minus1,

  653.                     isBE(dstFormat));

  654.         } else

  655.             fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);

  656.     }

  657. 

  658. #if HAVE_MMXEXT_INLINE

  659.     if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)

  660.         __asm__ volatile ("sfence" ::: "memory");

  661. #endif

  662.     emms_c();

  663. 

  664.     /* store changed local vars back in the context */

  665.     c->dstY         = dstY;

  666.     c->lumBufIndex  = lumBufIndex;

  667.     c->chrBufIndex  = chrBufIndex;

  668.     c->lastInLumBuf = lastInLumBuf;

  669.     c->lastInChrBuf = lastInChrBuf;

  670. 

  671.     return dstY - lastDstY;

  672. }

  673. 

  674. av_cold void ff_sws_init_range_convert(SwsContext *c)

  675. {

  676.     c->lumConvertRange = NULL;

  677.     c->chrConvertRange = NULL;

  678.     if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {

  679.         if (c->dstBpc <= 14) {

  680.             if (c->srcRange) {

  681.                 c->lumConvertRange = lumRangeFromJpeg_c;

  682.                 c->chrConvertRange = chrRangeFromJpeg_c;

  683.             } else {

  684.                 c->lumConvertRange = lumRangeToJpeg_c;

  685.                 c->chrConvertRange = chrRangeToJpeg_c;

  686.             }

  687.         } else {

  688.             if (c->srcRange) {

  689.                 c->lumConvertRange = lumRangeFromJpeg16_c;

  690.                 c->chrConvertRange = chrRangeFromJpeg16_c;

  691.             } else {

  692.                 c->lumConvertRange = lumRangeToJpeg16_c;

  693.                 c->chrConvertRange = chrRangeToJpeg16_c;

  694.             }

  695.         }

  696.     }

  697. }

  698. 

  699. static av_cold void sws_init_swscale(SwsContext *c)

  700. {

  701.     enum AVPixelFormat srcFormat = c->srcFormat;

  702. 

  703.     ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,

  704.                              &c->yuv2nv12cX, &c->yuv2packed1,

  705.                              &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);

  706. 

  707.     ff_sws_init_input_funcs(c);

  708. 

  709. 

  710.     if (c->srcBpc == 8) {

  711.         if (c->dstBpc <= 14) {

  712.             c->hyScale = c->hcScale = hScale8To15_c;

  713.             if (c->flags & SWS_FAST_BILINEAR) {

  714.                 c->hyscale_fast = ff_hyscale_fast_c;

  715.                 c->hcscale_fast = ff_hcscale_fast_c;

  716.             }

  717.         } else {

  718.             c->hyScale = c->hcScale = hScale8To19_c;

  719.         }

  720.     } else {

  721.         c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c

  722.                                                  : hScale16To15_c;

  723.     }

  724. 

  725.     ff_sws_init_range_convert(c);

  726. 

  727.     if (!(isGray(srcFormat) || isGray(c->dstFormat) ||

  728.           srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))

  729.         c->needs_hcscale = 1;

  730. }

  731. 

  732. SwsFunc ff_getSwsFunc(SwsContext *c)

  733. {

  734.     sws_init_swscale(c);

  735. 

  736.     if (ARCH_PPC)

  737.         ff_sws_init_swscale_ppc(c);

  738.     if (ARCH_X86)

  739.         ff_sws_init_swscale_x86(c);

  740. 

  741.     return swscale;

  742. }

  743. 

  744. static void reset_ptr(const uint8_t *src[], int format)

  745. {

  746.     if (!isALPHA(format))

  747.         src[3] = NULL;

  748.     if (!isPlanar(format)) {

  749.         src[3] = src[2] = NULL;

  750. 

  751.         if (!usePal(format))

  752.             src[1] = NULL;

  753.     }

  754. }

  755. 

  756. static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt,

  757.                                 const int linesizes[4])

  758. {

  759.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);

  760.     int i;

  761. 

  762.     for (i = 0; i < 4; i++) {

  763.         int plane = desc->comp[i].plane;

  764.         if (!data[plane] || !linesizes[plane])

  765.             return 0;

  766.     }

  767. 

  768.     return 1;

  769. }

  770. 

  771. static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst,

  772.                          const uint16_t *src, int stride, int h)

  773. {

  774.     int xp,yp;

  775.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);

  776. 

  777.     for (yp=0; yp<h; yp++) {

  778.         for (xp=0; xp+2<stride; xp+=3) {

  779.             int x, y, z, r, g, b;

  780. 

  781.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  782.                 x = AV_RB16(src + xp + 0);

  783.                 y = AV_RB16(src + xp + 1);

  784.                 z = AV_RB16(src + xp + 2);

  785.             } else {

  786.                 x = AV_RL16(src + xp + 0);

  787.                 y = AV_RL16(src + xp + 1);

  788.                 z = AV_RL16(src + xp + 2);

  789.             }

  790. 

  791.             x = c->xyzgamma[x>>4];

  792.             y = c->xyzgamma[y>>4];

  793.             z = c->xyzgamma[z>>4];

  794. 

  795.             // convert from XYZlinear to sRGBlinear

  796.             r = c->xyz2rgb_matrix[0][0] * x +

  797.                 c->xyz2rgb_matrix[0][1] * y +

  798.                 c->xyz2rgb_matrix[0][2] * z >> 12;

  799.             g = c->xyz2rgb_matrix[1][0] * x +

  800.                 c->xyz2rgb_matrix[1][1] * y +

  801.                 c->xyz2rgb_matrix[1][2] * z >> 12;

  802.             b = c->xyz2rgb_matrix[2][0] * x +

  803.                 c->xyz2rgb_matrix[2][1] * y +

  804.                 c->xyz2rgb_matrix[2][2] * z >> 12;

  805. 

  806.             // limit values to 12-bit depth

  807.             r = av_clip_c(r,0,4095);

  808.             g = av_clip_c(g,0,4095);

  809.             b = av_clip_c(b,0,4095);

  810. 

  811.             // convert from sRGBlinear to RGB and scale from 12bit to 16bit

  812.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  813.                 AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4);

  814.                 AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4);

  815.                 AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4);

  816.             } else {

  817.                 AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4);

  818.                 AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4);

  819.                 AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4);

  820.             }

  821.         }

  822.         src += stride;

  823.         dst += stride;

  824.     }

  825. }

  826. 

  827. static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst,

  828.                          const uint16_t *src, int stride, int h)

  829. {

  830.     int xp,yp;

  831.     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat);

  832. 

  833.     for (yp=0; yp<h; yp++) {

  834.         for (xp=0; xp+2<stride; xp+=3) {

  835.             int x, y, z, r, g, b;

  836. 

  837.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  838.                 r = AV_RB16(src + xp + 0);

  839.                 g = AV_RB16(src + xp + 1);

  840.                 b = AV_RB16(src + xp + 2);

  841.             } else {

  842.                 r = AV_RL16(src + xp + 0);

  843.                 g = AV_RL16(src + xp + 1);

  844.                 b = AV_RL16(src + xp + 2);

  845.             }

  846. 

  847.             r = c->rgbgammainv[r>>4];

  848.             g = c->rgbgammainv[g>>4];

  849.             b = c->rgbgammainv[b>>4];

  850. 

  851.             // convert from sRGBlinear to XYZlinear

  852.             x = c->rgb2xyz_matrix[0][0] * r +

  853.                 c->rgb2xyz_matrix[0][1] * g +

  854.                 c->rgb2xyz_matrix[0][2] * b >> 12;

  855.             y = c->rgb2xyz_matrix[1][0] * r +

  856.                 c->rgb2xyz_matrix[1][1] * g +

  857.                 c->rgb2xyz_matrix[1][2] * b >> 12;

  858.             z = c->rgb2xyz_matrix[2][0] * r +

  859.                 c->rgb2xyz_matrix[2][1] * g +

  860.                 c->rgb2xyz_matrix[2][2] * b >> 12;

  861. 

  862.             // limit values to 12-bit depth

  863.             x = av_clip_c(x,0,4095);

  864.             y = av_clip_c(y,0,4095);

  865.             z = av_clip_c(z,0,4095);

  866. 

  867.             // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit

  868.             if (desc->flags & AV_PIX_FMT_FLAG_BE) {

  869.                 AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4);

  870.                 AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4);

  871.                 AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4);

  872.             } else {

  873.                 AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4);

  874.                 AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4);

  875.                 AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4);

  876.             }

  877.         }

  878.         src += stride;

  879.         dst += stride;

  880.     }

  881. }

  882. 

  883. /**

  884.  * swscale wrapper, so we don't need to export the SwsContext.

  885.  * Assumes planar YUV to be in YUV order instead of YVU.

  886.  */

  887. int attribute_align_arg sws_scale(struct SwsContext *c,

  888.                                   const uint8_t * const srcSlice[],

  889.                                   const int srcStride[], int srcSliceY,

  890.                                   int srcSliceH, uint8_t *const dst[],

  891.                                   const int dstStride[])

  892. {

  893.     int i, ret;

  894.     const uint8_t *src2[4];

  895.     uint8_t *dst2[4];

  896.     uint8_t *rgb0_tmp = NULL;

  897. 

  898.     if (!srcStride || !dstStride || !dst || !srcSlice) {

  899.         av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n");

  900.         return 0;

  901.     }

  902.     memcpy(src2, srcSlice, sizeof(src2));

  903.     memcpy(dst2, dst, sizeof(dst2));

  904. 

  905.     // do not mess up sliceDir if we have a "trailing" 0-size slice

  906.     if (srcSliceH == 0)

  907.         return 0;

  908. 

  909.     if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) {

  910.         av_log(c, AV_LOG_ERROR, "bad src image pointers\n");

  911.         return 0;

  912.     }

  913.     if (!check_image_pointers((const uint8_t* const*)dst, c->dstFormat, dstStride)) {

  914.         av_log(c, AV_LOG_ERROR, "bad dst image pointers\n");

  915.         return 0;

  916.     }

  917. 

  918.     if (c->sliceDir == 0 && srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) {

  919.         av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n");

  920.         return 0;

  921.     }

  922.     if (c->sliceDir == 0) {

  923.         if (srcSliceY == 0) c->sliceDir = 1; else c->sliceDir = -1;

  924.     }

  925. 

  926.     if (usePal(c->srcFormat)) {

  927.         for (i = 0; i < 256; i++) {

  928.             int r, g, b, y, u, v, a = 0xff;

  929.             if (c->srcFormat == AV_PIX_FMT_PAL8) {

  930.                 uint32_t p = ((const uint32_t *)(srcSlice[1]))[i];

  931.                 a = (p >> 24) & 0xFF;

  932.                 r = (p >> 16) & 0xFF;

  933.                 g = (p >>  8) & 0xFF;

  934.                 b =  p        & 0xFF;

  935.             } else if (c->srcFormat == AV_PIX_FMT_RGB8) {

  936.                 r = ( i >> 5     ) * 36;

  937.                 g = ((i >> 2) & 7) * 36;

  938.                 b = ( i       & 3) * 85;

  939.             } else if (c->srcFormat == AV_PIX_FMT_BGR8) {

  940.                 b = ( i >> 6     ) * 85;

  941.                 g = ((i >> 3) & 7) * 36;

  942.                 r = ( i       & 7) * 36;

  943.             } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) {

  944.                 r = ( i >> 3     ) * 255;

  945.                 g = ((i >> 1) & 3) * 85;

  946.                 b = ( i       & 1) * 255;

  947.             } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) {

  948.                 r = g = b = i;

  949.             } else {

  950.                 av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE);

  951.                 b = ( i >> 3     ) * 255;

  952.                 g = ((i >> 1) & 3) * 85;

  953.                 r = ( i       & 1) * 255;

  954.             }

  955. #define RGB2YUV_SHIFT 15

  956. #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  957. #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  958. #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  959. #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  960. #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  961. #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  962. #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  963. #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  964. #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5))

  965. 

  966.             y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);

  967.             u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);

  968.             v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT);

  969.             c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24);

  970. 

  971.             switch (c->dstFormat) {

  972.             case AV_PIX_FMT_BGR32:

  973. #if !HAVE_BIGENDIAN

  974.             case AV_PIX_FMT_RGB24:

  975. #endif

  976.                 c->pal_rgb[i]=  r + (g<<8) + (b<<16) + ((unsigned)a<<24);

  977.                 break;

  978.             case AV_PIX_FMT_BGR32_1:

  979. #if HAVE_BIGENDIAN

  980.             case AV_PIX_FMT_BGR24:

  981. #endif

  982.                 c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24);

  983.                 break;

  984.             case AV_PIX_FMT_RGB32_1:

  985. #if HAVE_BIGENDIAN

  986.             case AV_PIX_FMT_RGB24:

  987. #endif

  988.                 c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24);

  989.                 break;

  990.             case AV_PIX_FMT_RGB32:

  991. #if !HAVE_BIGENDIAN

  992.             case AV_PIX_FMT_BGR24:

  993. #endif

  994.             default:

  995.                 c->pal_rgb[i]=  b + (g<<8) + (r<<16) + ((unsigned)a<<24);

  996.             }

  997.         }

  998.     }

  999. 

  1000.     if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) {

  1001.         uint8_t *base;

  1002.         int x,y;

  1003.         rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);

  1004.         if (!rgb0_tmp)

  1005.             return AVERROR(ENOMEM);

  1006. 

  1007.         base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;

  1008.         for (y=0; y<srcSliceH; y++){

  1009.             memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW);

  1010.             for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) {

  1011.                 base[ srcStride[0]*y + x] = 0xFF;

  1012.             }

  1013.         }

  1014.         src2[0] = base;

  1015.     }

  1016. 

  1017.     if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {

  1018.         uint8_t *base;

  1019.         rgb0_tmp = av_malloc(FFABS(srcStride[0]) * srcSliceH + 32);

  1020.         if (!rgb0_tmp)

  1021.             return AVERROR(ENOMEM);

  1022. 

  1023.         base = srcStride[0] < 0 ? rgb0_tmp - srcStride[0] * (srcSliceH-1) : rgb0_tmp;

  1024. 

  1025.         xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH);

  1026.         src2[0] = base;

  1027.     }

  1028. 

  1029.     if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0])

  1030.         for (i = 0; i < 4; i++)

  1031.             memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));

  1032. 

  1033. 

  1034.     // copy strides, so they can safely be modified

  1035.     if (c->sliceDir == 1) {

  1036.         // slices go from top to bottom

  1037.         int srcStride2[4] = { srcStride[0], srcStride[1], srcStride[2],

  1038.                               srcStride[3] };

  1039.         int dstStride2[4] = { dstStride[0], dstStride[1], dstStride[2],

  1040.                               dstStride[3] };

  1041. 

  1042.         reset_ptr(src2, c->srcFormat);

  1043.         reset_ptr((void*)dst2, c->dstFormat);

  1044. 

  1045.         /* reset slice direction at end of frame */

  1046.         if (srcSliceY + srcSliceH == c->srcH)

  1047.             c->sliceDir = 0;

  1048. 

  1049.         ret = c->swscale(c, src2, srcStride2, srcSliceY, srcSliceH, dst2,

  1050.                           dstStride2);

  1051.     } else {

  1052.         // slices go from bottom to top => we flip the image internally

  1053.         int srcStride2[4] = { -srcStride[0], -srcStride[1], -srcStride[2],

  1054.                               -srcStride[3] };

  1055.         int dstStride2[4] = { -dstStride[0], -dstStride[1], -dstStride[2],

  1056.                               -dstStride[3] };

  1057. 

  1058.         src2[0] += (srcSliceH - 1) * srcStride[0];

  1059.         if (!usePal(c->srcFormat))

  1060.             src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1];

  1061.         src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2];

  1062.         src2[3] += (srcSliceH - 1) * srcStride[3];

  1063.         dst2[0] += ( c->dstH                         - 1) * dstStride[0];

  1064.         dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1];

  1065.         dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2];

  1066.         dst2[3] += ( c->dstH                         - 1) * dstStride[3];

  1067. 

  1068.         reset_ptr(src2, c->srcFormat);

  1069.         reset_ptr((void*)dst2, c->dstFormat);

  1070. 

  1071.         /* reset slice direction at end of frame */

  1072.         if (!srcSliceY)

  1073.             c->sliceDir = 0;

  1074. 

  1075.         ret = c->swscale(c, src2, srcStride2, c->srcH-srcSliceY-srcSliceH,

  1076.                           srcSliceH, dst2, dstStride2);

  1077.     }

  1078. 

  1079. 

  1080.     if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) {

  1081.         /* replace on the same data */

  1082.         rgb48Toxyz12(c, (uint16_t*)dst2[0], (const uint16_t*)dst2[0], dstStride[0]/2, ret);

  1083.     }

  1084. 

  1085.     av_free(rgb0_tmp);

  1086.     return ret;

  1087. }